kth.sePublications
Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Neural Network Quantum State Ansatz for the Nuclear Pairing Problem
KTH, School of Engineering Sciences (SCI), Physics.
2024 (English)Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesisAlternative title
Neuralt Nätverks Kvanttillståndsansats för Kärnparsproblemet (Swedish)
Abstract [en]

As a degree project in Theoretical Physics, the variational MCMC-scheme aided by neural network quantum states was examined for the purpose ofsolving the nuclear pairing model. The method entailed minimization of the local energy sampled via the Born distribution obtained through the neural network output.Both the ground and excited states' energies were computed, where the latter case used an extended loss function which included the overlap to the former.The NNQS-ansatz worked well when emulating the ground state, in which case the Stochastic Reconfiguration optimization method was particularly effective. This optimization method resulted in relative fast convergence to low variance states, and did not require a large number of hyperparameter modifications. Ultimately, all resulting energy intervals encompassed the exact ground state solutions, and had relative errors equal to or near zero.For the excited states, the VMC-NNQS was less effective, as each individual occupation number state investigated required considerable hyperparameter testing before reasonably low lying energy eigenstates could be obtained. Moreover, the convergence properties were less distinguished than for the ground state, as the optimization struggled to maintain orthogonality to the ground state. Nonetheless, the final results included the nearest solutions of the first excited states for several systems and indicated correlation energies similar to those of the ground state.

Abstract [sv]

Som examensarbete inom teoretisk fysik undersöktes den variationella MCMC-metoden tillsammans med neurala nätverk i syfte att lösa kärnparsmodellen. Metoden innebar minimering av den lokala energin som samplades via Born-fördelningen erhållen genom utdata från neurala nätverksapproximationer. Både grundtillståndets och exciterade tillstånds energier beräknades, där det senare fallet använde en utökad kostnadsfunktion som inkluderade överlappet med det förnämnda. NNQS-ansatsen fungerade väl vid emulering av grundtillståndet, i vilket fall optimeringsmethoden stokastisk omkonfigurering (Stochastic Reconfiguration) var särskilt effektivt. Denna optimeringsmetod resulterade i relativt snabb konvergens till lågvarianstillstånd och krävde inte ett stort antal hyperparametriska modifieringar. De slutliga energiintervallen innefattade de exakta lösningarna för grundtillstånden med en relativ felmarginal lika med eller nära noll. För exciterade tillstånd var VMC-NNQS mindre effektivt, eftersom varje enskilt ockupationstillstånd som undersöktes krävde en ansenlig mängd hyperparametrisk testning innan rimligt låga egentillstånd kunde erhållas. Dessutom var konvergensensegenskaperna mycket mindre särspäglade än för grundtillståndet, eftersom optimeringen inte fullt kunde upprätthålla ortogonaliteten mot grundtillståndet. Likväl inkluderade de slutliga resultaten de närmaste lösningarna av de första exciterade energierna för ett flertal system, och visade på korrelationsenergier liknande de för grundtillståndet.

Place, publisher, year, edition, pages
2024.
Series
TRITA-SCI-GRU ; 2024:048
Keywords [en]
Neural Network, NNQS, Variational Monte Carlo, VMC, Quantum Many-Body Problem
Keywords [sv]
Neurala nätverk, NNQS, Variations Monte Carlo, VMC, Kvantfysik, Kvanttillstånd
National Category
Physical Sciences
Identifiers
URN: urn:nbn:se:kth:diva-346486OAI: oai:DiVA.org:kth-346486DiVA, id: diva2:1858209
Subject / course
Physics
Educational program
Master of Science - Engineering Physics
Supervisors
Examiners
Available from: 2024-05-16 Created: 2024-05-16 Last updated: 2024-05-16Bibliographically approved

Open Access in DiVA

fulltext(3760 kB)148 downloads
File information
File name FULLTEXT01.pdfFile size 3760 kBChecksum SHA-512
c0bf56005ae5688e690bf13c21f2461f5785a4aa507f227c726ebee0ed022631d59d48ae0ddd52368149d46ebf26266669476f0b200dad9d077dd6331f890d38
Type fulltextMimetype application/pdf

By organisation
Physics
Physical Sciences

Search outside of DiVA

GoogleGoogle Scholar
Total: 148 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

urn-nbn

Altmetric score

urn-nbn
Total: 359 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf